Representation of non-spatial and spatial information in ...Representation of non-spatial and spatial information in ... like” memory in animals – memory that requires an integration

BEHAVIORAL NEUROSCIENCEORIGINAL RESEARCH ARTICLE

published: 28 October 2011doi: 10.3389/fnbeh.2011.00069

Representation of non-spatial and spatial information inthe lateral entorhinal cortexSachin S. Deshmukh1,2 and James J. Knierim1,2,3*

1 Krieger Mind/Brain Institute, Johns Hopkins University, Baltimore, MD, USA2 Department of Neurobiology and Anatomy, University of Texas Medical School at Houston, Houston, TX, USA3 Solomon H. Snyder Department of Neuroscience, Johns Hopkins University School of Medicine, Baltimore, MD, USA

Edited by:

Donald A. Wilson, New YorkUniversity School of Medicine, USA

Reviewed by:

Rebecca D. Burwell, BrownUniversity, USAAnne-Marie Mouly, CNRS-Universitéde Lyon, UMR 5020, France

*Correspondence:

James J. Knierim, Krieger Mind/BrainInstitute, Johns Hopkins University,338 Krieger Hall, 3400 North CharlesStreet, Baltimore, MD 21218, USA.e-mail: [email protected]

Some theories of memory propose that the hippocampus integrates the individual itemsand events of experience within a contextual or spatial framework. The hippocampusreceives cortical input from two major pathways: the medial entorhinal cortex (MEC) andthe lateral entorhinal cortex (LEC). During exploration in an open field, the firing fields ofMEC grid cells form a periodically repeating, triangular array. In contrast, LEC neurons showlittle spatial selectivity, and it has been proposed that the LEC may provide non-spatial inputto the hippocampus. Here, we recorded MEC and LEC neurons while rats explored an openfield that contained discrete objects. LEC cells fired selectively at locations relative to theobjects, whereas MEC cells were weakly influenced by the objects. These results providethe first direct demonstration of a double dissociation between LEC and MEC inputs tothe hippocampus under conditions of exploration typically used to study hippocampal placecells.

Keywords: hippocampus, objects, navigation, memory, medial entorhinal cortex, grid cells

INTRODUCTIONThe hippocampus is critically involved in episodic memory inhumans (Scoville and Milner, 1957; O’Keefe and Nadel, 1978;Vargha-Khadem et al., 1997; Squire et al., 2004) and “episodic-like” memory in animals – memory that requires an integrationof the what, where, and when components of a memory (Claytonand Dickinson, 1998; Eichenbaum and Fortin, 2005). The mostsalient firing correlate of hippocampal neurons in freely movingrats is the location of the animal, which led to the notion thatthe hippocampus provides a spatial framework to organize andinterrelate the items and events of experience and to allow flexi-ble memory storage (O’Keefe and Nadel, 1978). Non-spatial inputsalso influence the activity of hippocampal cells (Wiener et al., 1989;Hampson et al., 1999; Wood et al., 1999; Rivard et al., 2004; Lenck-Santini et al., 2005), often by modulating the underlying placefield of the cell (O’Keefe, 1976; Moita et al., 2003; Komorowskiet al., 2009; Manns and Eichenbaum, 2009). The creation of suchcontext-specific, “item + place” conjunctive representations maybe the key contribution of the hippocampus to episodic memory.

Understanding the computations involved in the creation ofhippocampal representations requires a detailed knowledge of theinformation that is encoded in its afferent structures. There are twomajor cortical inputs to the hippocampus, the medial entorhinalcortex (MEC) and the lateral entorhinal cortex (LEC). The MECconveys highly specific spatial information to the hippocampus inthe form of grid cells, boundary cells, and head direction cells. Gridcells fire in multiple locations as a rat explores an environment,and the firing locations are arranged as the vertices of an exquis-itely regular grid of equilateral triangles that tessellate the floorof the environment (Hafting et al., 2005). A subset of grid cellsare also modulated by the head direction of the animal (Sargolini

et al., 2006). Boundary cells fire when the rat is located near a wallor edge of an environment, and they may provide information toalign the grid cells to the boundaries of an apparatus (Savelli et al.,2008; Solstad et al., 2008). Anatomically, the MEC is connectedstrongly to other regions that demonstrate similar spatial firingproperties, such as the subiculum, presubiculum, parasubiculum,and retrosplenial cortex (Chen et al., 1994; Taube, 1995; Sharp,1997; Cho and Sharp, 2001; Hargreaves et al., 2007; Lever et al.,2009; Boccara et al., 2010; Knierim and Hamilton, 2011).

In contrast to the MEC, neurons in the LEC do not show strongspatial firing (Hargreaves et al., 2005; Yoganarasimha et al., 2010)or movement-related theta (Deshmukh et al., 2010). Some studieshave demonstrated that LEC neurons respond to individual items,such as odors, pictures of objects, or views of three-dimensionalobjects (Zhu et al., 1995a,b; Young et al., 1997; Wan et al., 1999).Anatomically, the LEC receives major input from the perirhinalcortex, which is involved in object-recognition and familiarity(Aggleton and Brown, 1999; Murray et al., 2007).

Based on anatomical arguments that the MEC is part of thebrain’s dorsal (“where”) processing stream and the LEC is partof the ventral (“what”) stream (Figure 1), a number of investi-gators have proposed that place cells in the hippocampus derivetheir spatial selectivity from their MEC input and their non-spatialmodulation from the LEC input (Suzuki et al., 1997; Burwell, 2000;Witter and Amaral, 2004; Knierim et al., 2006; Manns and Eichen-baum, 2006). However, although prior studies of LEC have shownresponsiveness to individual items, these studies took place underconditions that were different from the conditions typically usedto study hippocampal place cells and they did not explicitly distin-guish MEC and LEC (Zhu et al., 1995a,b; Young et al., 1997; Wanet al., 1999), limiting the ability to compare LEC firing directly

Frontiers in Behavioral Neuroscience www.frontiersin.org October 2011 | Volume 5 | Article 69 | 1

http://www.frontiersin.org/Behavioral_Neuroscience

http://www.frontiersin.org/Behavioral_Neuroscience/editorialboard



http://www.frontiersin.org/Behavioral_Neuroscience/about

http://www.frontiersin.org/Behavioral_Neuroscience/10.3389/fnbeh.2011.00069/abstract

http://www.frontiersin.org/Community/WhosWhoDetails.aspx?UID=38877&d=1&sname=SachinDeshmukh&name=Science

http://www.frontiersin.org/Community/WhosWhoDetails.aspx?UID=11727&d=0&sname=JamesKnierim&name=all people

mailto:[email protected]


http://www.frontiersin.org

http://www.frontiersin.org/Behavioral_Neuroscience/archive

Deshmukh and Knierim Objects and space in LEC

FIGURE 1 | Anatomical segregation of cortical inputs to hippocampus

(Burwell, 2000;Witter and Amaral, 2004). The LEC receives major inputfrom the perirhinal cortex, part of the brain’s ventral (“what”) pathway. TheMEC receives input from the postrhinal (parahippocampal) cortex, part ofthe dorsal (“where”) pathway. The MEC also receives major spatial inputsfrom the presubiculum, postsubiculum, and retrosplenial cortex, all ofwhich show stronger spatial tuning than the postrhinal cortex (Knierim,2006). The projections of LEC and MEC to CA1 remain segregated alongthe transverse (proximal–distal) axis of the hippocampus, whereas theprojections to the DG and CA3 converge onto the same anatomical regions.(For simplicity, a number of anatomical connections have been excludedfrom this diagram. super: superficial layers II and III, the inputs to thehippocampus. deep: deep layers V and VI, which receive feedback from thehippocampus.)

with place cell activity and MEC grid cell activity. The presentstudy thus investigated the responses of MEC and LEC neuronswhile freely moving rats foraged for food in an open field thatcontained a number of discrete objects, to test whether LEC andMEC neurons differentially conveyed information about objectsand spatial location to the hippocampus. We report here that LECcells were more strongly influenced by local objects than MECcells, firing preferentially at the locations of objects as well as atlocations at a distance from the objects.

MATERIALS AND METHODSANIMALS AND SURGERYSeven male Long–Evans rats, aged 5–6 months, were housed indi-vidually on a 12:12-h reversed light–dark cycle. All experimentswere performed in the dark portion of the cycle. Animal care, sur-gical procedures, and euthanasia were performed in accordancewith the National Institutes of Health (NIH) and the Universityof Texas Health Sciences Center at Houston Institutional AnimalCare and Use Committee (IACUC) guidelines.

Under surgical anesthesia, 2 animals were implanted with ahyperdrive with 18 tetrodes directed at the LEC of the right hemi-sphere, and 5 rats were implanted with a hyperdrive with 9 tetrodeseach directed at LEC and MEC (MEC units were recorded from 3of these rats). LEC bundles were centered at 7.7–8.1 mm posteriorto bregma and 3.2–3.5 mm lateral to midline. The electrodes wereangled at 25˚ medio-laterally to allow them to access the lateral tomedial extent of LEC. MEC bundles were positioned with the mostposterior tetrode at 600–800 μm anterior to the transverse sinusand 4.8–5 mm lateral to the midline. The MEC tetrodes traveledvertically. This allowed us to record from neurons along the dorsalto ventral axis of MEC.

TRAINING AND EXPERIMENTAL PROTOCOLRats were allowed to recover for 5–6 days after surgery. During thetraining and recording sessions the rats were maintained at 80–90% of their free feeding weights. The rats were trained to foragefor irregularly distributed chocolate sprinkles in a 1.2-m × 1.5-mbox with 0.3 m high walls. The box had 34 irregularly distributedanchoring positions where objects could be placed. No objectswere present during the training sessions. The box was placed 6′′above the floor in a room with numerous visual cues, such asother behavioral apparatus, open curtains, and doors (Figure 2).Once the rats were trained to forage for six consecutive 15-minsessions with the preamplifier headstages plugged in, and the elec-trodes were deemed to be in the target regions (Deshmukh et al.,2010), the object-related recording sessions commenced. The firstsession of the first day of recording was identical to the train-ing sessions without objects, but sessions 2–6 had four objectsin a configuration that was fixed for each rat. This is referred toas the standard object configuration. All six sessions on the sec-ond day had objects in the standard configuration. Starting onday 3, two object-manipulation sessions (sessions 3 and 5) wereinterspersed with standard object configuration sessions. Object-manipulations consisted of either (a) introducing a novel objector (b) moving one (or sometimes, two) objects from their stan-dard location. Sessions 3 and 5 alternated as novel and misplacedobject sessions on consecutive days, to minimize order effects. Dif-ferent object configurations served as standard configurations fordifferent rats, but the same objects and their spatial configurationserved as the standard on all days of recording for a given rat.For two of the rats, an additional (seventh) session was run in asimilar box in an adjacent room with numerous visual cues, butno objects in the box. One of the rats failed to run six consecu-tive sessions on most days. The protocol was modified to includeonly one manipulation session (session 3) each day for this rat.Even for other rats, if they showed a tendency to forage poorlyin a session or two, the recording was stopped before completionof the six-session sequence. The sessions that were subjectivelyjudged to have poor coverage of the box were excluded fromanalysis, without regard to the activities of the neurons recordedin these sessions. The box was cleaned with 70% ethanol at theend of the day’s recording sequence for each rat. After training,the rats did not usually defecate or urinate in the box duringforaging. In the rare instances in which urination or defecationoccurred, the feces or urine was removed immediately, and thearea wiped down to spread the odor over a substantial portion(more than 1/4th) of the box. The approximate position and timeof these events were noted. After session 1, units were quicklyassessed to determine which objects to manipulate in the mis-placed object session (i.e., to move an object for which at leastone cell fired in session 1). Thus there was a 30- to 45-min gapbetween the first and the second sessions. All other sessions hada variable 5–7 min interval. The rats had access to water betweensessions.

Tetrodes were advanced at least 100 μm while listening to thechanging activity starting on the third day of recording to sampledifferent units on different days. On most days, there was at least a16-h gap between the time when tetrodes were last moved and thestart of recordings. On some days, when no neurons were present






FIGURE 2 | Experimental protocol. (A) Recording environment. (B)

Representative objects used in the experiments. (C) Typicalexperimental protocol. Rats foraged for chocolate sprinkles for 15 mineach in six consecutive sessions in the presence of objects. Sessions 3and 5 were object-manipulation sessions in which either a novel object

was introduced (session 3, here) or one of the objects was misplaced(session 5 here); the type of object-manipulation was counterbalancedbetween sessions 3 and 5 across days. Session 7, in which ratsexplored a box in a different room with no objects, was run in only thelast two rats.

at the start of the day, some tetrodes were moved a little (20–200 μm) until units were obtained. On these days, there was at leasta 4-h delay between moving the tetrodes and the recording session.Recordings were terminated when all tetrodes were deemed to bein layer I of the cortex using functional landmarks described pre-viously (Deshmukh et al., 2010). The number of recording daysin all but one rat ranged from 12–14 (median: 12 days, outlier:6 days).

ObjectsObjects used in this experiment were mostly small toys and had avariety of textures, shapes, colors, and sizes (Figure 2). The smallestdimension for any object was 2.5 cm, and the largest was 15 cm.

RECORDING HARDWARETetrodes were made from either 12.5 μm nichrome wire or 17 μm90% platinum- 10% iridium wire (California Fine Wire, GroverBeach, CA, USA). Nichrome tetrodes were gold plated to bringtheir impedance down to approximately 200 kΩ. Pt–Ir wires werenot plated, and their impedances were approximately 700 kΩ.Recordings were performed with the Cheetah Data AcquisitionSystem (Neuralynx, Bozeman, MT, USA) as described previously(Deshmukh et al., 2010).

DATA ANALYSISUnit isolationManual cluster cutting with custom software was used to isolateactivities of single units (Deshmukh et al., 2010). Each cell was

assigned an isolation quality score on a subjective scale of 1 (verywell-isolated) to 5 (poorly isolated), based on how well the clusterwas separated from the neighboring clusters and the background.Spatial firing characteristics of the units were not used for assign-ing isolation quality. Clusters with quality of 3 or better, firing atleast 50 spikes in a session, were used for the subsequent analyses.Fast spiking cells with mean firing rates >10 Hz were assumed tobe interneurons and were excluded from the analysis (Frank et al.,2001; Hargreaves et al., 2005).

Firing rate mapThe position and head direction of the rat were recorded usingLEDs connected to the hyperdrive and an overhead camera (Model1300, Cohu Inc., San Diego, CA, USA). The area of the box wassegmented into 3.4-cm square bins. The firing rate map of eachcell was constructed by dividing the number of spikes in eachbin with the amount of time the rat spent in that bin. Theseunsmoothed rate maps were used for object response calcula-tions. The rate maps were smoothed using the adaptive binningalgorithm described by Skaggs et al. (1996), for use in spatialinformation score calculations and for illustrations.

Spatial informationSpatial tuning of single units was quantified using the spatial infor-mation measure devised by Skaggs et al. (1996), which quantifiesthe amount of information (in bits) about the rat’s location con-veyed by a single spike. The probability of obtaining a spatialinformation score for a given unit by chance was estimated using






a shuffling procedure. The neuron’s spike train was shifted intime with respect to the rat’s trajectory with 1000 random timelags (minimum shift of 30 s), and the spatial information wascalculated on the rate maps calculated for each random shift.The probability of obtaining the observed information score bychance is the fraction of randomly time shifted trials having spa-tial information equal to or greater than the observed informationscore. We used a significance threshold of p < 0.01 for the spatialinformation.

Object-responsivenessObjects remained fixed at their spatial locations throughout a ses-sion, thus making it possible to estimate object-related firing of agiven unit by comparing its firing rate when the rat occupied pixelsnear the objects with its firing rate when the rat occupied pixelsaway from the objects. Since spatial location of the objects wasused as a surrogate for the objects in this analysis, only the cellswith spatial information scores >0.25 bits/spike and probabilityof obtaining the information score by chance <0.01 were usedfor the purpose, unless specified otherwise. This selection processallowed the removal of cells that fired diffusely or unreliably in theenvironment.

To measure the relative firing rate of each neuron when therat occupied locations close to, vs. away from, the objects, anobject-responsiveness index (ORI) was calculated with the equa-tion (On − A)/(On + A), where On is the mean of all firing ratesin pixels within a 5-pixel (3.4 cm/pixel) radius of object n and A isthe mean of all firing rates in pixels that were more than 5 pixelsfrom all objects. Unsmoothed firing rate maps were used in thesecalculations. For sessions with the standard object configuration,the ORI was calculated five times for each neuron recorded in thesession; that is, the ORI was calculated for each of the four objectsindividually (i.e., On was the firing rate around object n only) aswell as a fifth measurement for all four-objects-together (i.e., On

was the mean firing rate for all pixels within a 5-pixel radius ofall four objects). The value of the largest of the five ORI mea-surements for each cell is referred to as ORImax in Figure 4A andcorresponding text.

To test the statistical significance of the ORI, the distribu-tion of ORIs that could be obtained by chance for a given unitwas estimated using 1000 random object placements on the ratemap, followed by ORI calculation for each random placement.For each simulated test run, four “objects” were located at randomlocations on the cell’s rate map, with the minimum and maxi-mum distances between objects set at 12 and 34 pixels respectively(these numbers correspond to the actual minimum and maximumdistances between standard objects in the seven rats in this exper-iment). Random object placements with less than 15 occupiedpixels around the objects were also excluded from the analysis.The five ORI calculations were performed on this simulated testrun. The simulation was repeated 1000 times, each with a differentset of random object locations, to generate an expected distribu-tion of ORI values that would occur by chance. The probabilityof obtaining the observed ORI [p(ORI)] is the fraction of ORIsfor the random object placement simulations that were equal toor more than the observed ORI. For example, if the observed ORIwas greater than every 1 of the 1000 simulated values, the p(ORI)

value would be 0.001. Each neuron generated five p(ORI) values(one for each ORI value), and the smallest of the five probabili-ties was referred to as the pmin(ORI). [Note that pmin(ORI) is notalways the probability of ORImax,due to differences in the shapes ofthe control distributions for single-object vs. four-objects-togetherconditions.]

Because the ORI value depends both on the firing rate of neu-rons near objects and the firing rate away from objects, comparisonbetween brain areas is tenuous if the regions differ in the firingrate away from objects (variable A in the ORI equation). However,because the p(ORI) measure is calculated with simulated data fromeach region, it effectively normalizes this confound. Thus, to com-pare the LEC to the MEC, we performed statistical tests on thepmin(ORI) values, rather than the ORI values themselves. χ2 withcorrection for continuity was used to analyze 2 × 2 contingencytables comparing proportions of putative object-responsive neu-rons in LEC and MEC. ORI and p(ORI) were also calculated formisplaced and novel objects in the misplaced and novel objectsessions and for the corresponding locations of these objects inthe standard object configuration sessions that flanked them. Forthese object-manipulation sessions, the ORI was calculated onlyfor individual objects, not for all objects together.

LEC place fields away from objectsWe observed place fields in LEC at some distance from objectsthat were stable across multiple sessions in the presence of objects.To objectively classify these units using session 1 rate maps, weused the following criteria: (1) the rate maps had to show a pixelby pixel correlation coefficient with the following session greaterthan 0.71, which was the mean correlation coefficient for the LECpopulation +1 SD; (2) the spatial information scores had to besignificant at p < 0.01, and higher than 0.4 bits/spike; and (3) thep(ORI) for the 4-objects-together had to be >0.4 (0.319 was themaximum value for the ORI p-value for the 4-objects-together forany neuron which showed object-responsiveness to a single objectin any standard session). These criteria identified neurons that hadstable spatial firing fields away from objects.

HISTOLOGYLocations of isolated single units were determined at the end of anexperiment by localizing tetrode tracks on coronal sections of therat brains, as described previously (Deshmukh et al., 2010).

RESULTSSELECTIVITY OF LEC AND MEC NEURONS IN THE PRESENCE OFOBJECTSMultiple single units were recorded from the superficial layers ofLEC and MEC (i.e., the layers that project to the hippocampus;Figure 1) while rats foraged for food reward in a 1.2-m × 1.5-mbox in the presence of objects (Figures 2A,B). On a typical day,four sessions in which the objects were placed in a standard con-figuration were interleaved with two sessions in which a subsetof the objects was moved to new locations or a novel object wasintroduced into the box (Figure 2C). Because the objects occupiedfixed positions in the standard sessions, a common measure of spa-tial information content (Skaggs et al., 1996) was used to initiallycharacterize the firing of LEC and MEC neurons in the first session






of the day. The distributions of spatial information scores from 75LEC and 49 MEC neurons were not statistically different underthese conditions (Figure 3A; LEC median = 0.34 bits/spike; MECmedian = 0.36 bits/spike; Wilcoxon rank sum test, p = 0.947). Thisresult is in stark contrast to previous studies that showed that, inthe absence of objects, MEC conveyed much more spatial infor-mation than LEC in both simple (Hargreaves et al., 2005) andcomplex (Yoganarasimha et al., 2010) environments. In particular,the distribution of information scores from LEC was qualita-tively different from the distributions reported in the previous

studies, with a skew toward high values that was absent in theprevious studies (Hargreaves et al., 2005; Yoganarasimha et al.,2010).

The stability of LEC and MEC rate maps between two consec-utive sessions with the standard object configuration (sessions 1and 2) was estimated using pixel by pixel correlation coefficientsbetween the two sessions for each cell. The distribution of thesecoefficients in LEC was not significantly different from that inMEC (LEC median = 0.53; MEC median = 0.56; Wilcoxon ranksum test, p = 0.479; Kolmogorov–Smirnov test, p = 0.916).

FIGURE 3 | Spatial information scores in LEC are comparable to those in

MEC in the presence of objects. (A) Distributions of spatial informationscores in LEC and MEC in session 1 with objects in their standardconfiguration. (B,C) Firing rate maps of LEC (B) and MEC (C) neurons withstatistically significant (p < 0.01) spatial information scores greater than

0.4 bits/spike in the first session. White circles mark locations of objects. Bluecorresponds to no firing while red corresponds to the peak firing rate.Numbers at the top of each rate map indicate peak firing rate (pk) in Hz andspatial information score (i) in bits/spike. Numbers at the left and right of thefigure indicate unit numbers.






Both the MEC and LEC histograms in Figure 3A showed adip around 0.4 bits/spike; the firing rate maps of neurons withspatial information scores higher than this value are shown inFigures 3B,C. Some LEC neurons fired selectively when the ratwas located close to one or more objects (e.g., units 7, 13, 14, 19,20, and 22) whereas other neurons fired selectively in locations at adistance from the objects, such as corners, walls, or locations nearthe center of the box (e.g., units 2, 8, 10, 12, 17, 21, and 23). In orderto test whether the increased LEC selectivity compared to previousstudies was an artifact of the recording conditions of the currentstudy, an additional session without objects was administered tothe last two rats (Figure 2C). To ensure that any spatial selectivityseen without objects was not confounded by potential effects ofthe animal’s memory of the objects, these additional sessions wereperformed in a different room. Figure A1 in Appendix shows thatthe spatial information scores of LEC neurons in the presence ofobjects were higher than the scores in the absence of objects, inagreement with prior reports that spatial information is low forLEC in the absence of objects. (Note that the overall decrease inspatial selectivity cannot be interpreted as similar to hippocampalplace field remapping between environments, as such remappingwould result in a combination of both increases and decreases inspatial selectivity across cells.)

As expected, MEC neurons showed a number of units withspatially selective responses (Figure 3C), and the distribution ofspatial information scores (Figure 3A) resembled the distributionsfrom previous studies (Hargreaves et al., 2005; Yoganarasimhaet al., 2010). (As with LEC, it was not possible to directly com-pare the spatial information scores between studies, as the spatialinformation measure is sensitive to differences between studies inthe size of the apparatus and in the size of the occupancy binsof the rate maps.) A number of MEC units were grid cells (Haft-ing et al., 2005; e.g., units 1, 4, 11, and 16) and boundary-relatedcells (Savelli et al., 2008; Solstad et al., 2008; e.g., units 6, 7, and10). Other cells fired in apparently arbitrary locations, both nearand away from the object locations (e.g., units 3 and 14), withno obvious tendency to concentrate firing near the objects (seebelow).

To quantify the responsiveness to objects, an ORI was definedas (On − A)/(On + A), where On is the mean firing rate within a17 cm (5-pixel) radius of object n and A is the mean firing rate ofall pixels outside the 5-pixel radius of all four objects. The ORI wasalso calculated for all objects together (i.e., On was the mean firingrate for all pixels within a 5-pixel radius of all four objects); thuseach cell produced five different ORI values (see Materials andMethods). Only neurons with statistically significant (p < 0.01)spatial information scores greater than 0.25 bits/spike were usedfor this and subsequent analyses. The distributions of the highestof the five ORI values for each cell (ORImax) in LEC and MEC insession 1 are shown in Figure 4Ai. The LEC distribution shows awider spread of ORImax than MEC. A direct statistical comparisonof ORImax between MEC and LEC is tenuous, as the differencesin the spatial firing characteristics of LEC and MEC neurons awayfrom the objects (variable A in the ORI equation) may affect thevalue of ORI. For example, because MEC cells are known to firein spatial patterns that do not depend on objects (e.g., grid cells inMEC), their firing rates away from objects might be higher than

LEC neurons and thereby decrease the magnitude of the ORI. Toaddress this problem, we calculated independently for each neuronthe probability that its ORI was due to chance [p(ORI)] using arandomization procedure (see Materials and Methods). Each cellgenerated five p(ORI) values (one for each of the five ORI cal-culations done per neuron), and the lowest p(ORI) was denotedpmin(ORI).The distribution of pmin(ORI) for LEC was wider thanMEC in session 1 (Figure 4Bi; Siegel–Tukey, p = 0.0018), althoughthere was no significant difference between the median valuesof the distributions (LEC median = 0.165, MEC median = 0.197;Wilcoxon rank sum, p = 0.6867). Moreover, LEC (11/41) had a sig-nificantly larger number of neurons with a pmin(ORI) < 0.05 thanMEC (1/28; χ2 = 4.75, one tailed p = 0.0148). Subsequent sessionswith objects in the standard configuration (Sessions 2, 4, and 6)showed a similar trend (Figure 4; Figures A2 and A3 in Appen-dix). These data indicate that LEC cells were more responsive toobjects than MEC cells.

RESPONSES TO OBJECT NOVELTY AND TRANSLOCATIONThe previous analyses were performed on the standard sessions,in which familiar objects were located in their standard positions,in order to measure baseline firing properties of the neurons rel-ative to the objects. Object-manipulation sessions, in which novelobjects were introduced or a standard object was translocated,were interleaved with the standard sessions. LEC cells often firedat multiple objects, and the firing was not always consistent acrosssessions. For example, unit 1 of Figure 5A fired at all 4 objectsin standard sessions 1, 2, 4, and 6, but the relative firing rates ateach object were variable. Moreover, in session 3, the cell fired at anovel object that was introduced into the environment, but firedonly very weakly at 2 of the 4 standard objects. When 2 objectswere moved in session 5, the cell fired strongly at one of the newobject locations (the top left corner of the box) and weakly at theother object location. The cell was silent at both of these loca-tions in the standard object sessions. To quantify the number ofcells that responded consistently to the objects across the first twostandard sessions, we considered whether each cell had the samepattern of significant pORI values across both sessions. For exam-ple, if a cell had a significant pORI for object 1, object 3, and allobjects combined (or any subset of the 5 ORI measures) in bothsessions, it was considered stable. Of the 15 cells that had at leastone significant pORI in either session, only 3 had the same pat-tern of significant pORI measures across the two sessions. Acrossall standard sessions, none of the 22 cells with a significant pORIin at least one session had the same pattern across all sessions.Thus, the pixel by pixel correlation coefficients between rate mapsreported earlier do not capture the variability in response to indi-vidual objects shown here, as the rate map correlations take intoaccount locations away from objects and are less sensitive to firingrate differences confined to small locations around the objects. Wedid not attempt to create rate map correlations using only the pix-els around objects, because the reduced number of pixels is likelyto make such measurements noisy and difficult to interpret ontheir own, in the absence of an absolute standard correlation scoreagainst which to measure response stability.

A few LEC units showed selective firing at a subset of the objects.Unit 2 was recorded simultaneously with unit 1 (on a separate






FIGURE 4 | Lateral entorhinal cortex (LEC) neurons display higher

object-responsiveness than MEC neurons. (A) Distribution of the ORImax ofLEC (top) and MEC (bottom) neurons in the four standard sessions. (B)

Distribution of the pmin(ORI). White bars indicate cells that showed statisticallysignificant (p < 0.05) object-related firing. LEC showed a significantly larger

proportion of neurons with object-related firing in three of the four standardsessions compared to MEC (session 1 LEC:11/41, MEC:1/28, χ2 = 4.75, onetailed p = 0.0148; session 2 LEC:14/43, MEC:5/36, χ2 = 3.74, one tailedp = 0.027; session 4 LEC: 7/31, MEC: 4/27, χ2 = 0.174, one tailed p = 0.308,n.s.; session 6 LEC 10/28, MEC 0/26, χ2 = 9.15, one tailed p = 0.0015).

tetrode). In contrast to unit 1, this cell fired strongly and consis-tently around only one of the objects, and it did not fire at eitherthe novel (session 3) or the misplaced (session 5) objects. Interest-ingly, in session 5, the cell continued to fire at the location wherethe lower left object had been in the standard sessions, actingmore like a “place cell” than an object cell (although it is possiblethat the presence of objects is required to generate this apparentspatial firing; see below). Rate maps for all neurons recorded inLEC and MEC are shown in Figures A2 and A3 in Appendix,respectively.

In LEC, 5 of the 29 neurons that met analysis criteria showedsignificantly higher firing at the novel object than regions awayfrom the novel and standard objects, which is a higher pro-portion than expected by chance at an alpha level of 0.05 (testfor proportions, z = 3.0, p = 0.0013). All 5 neurons also showedpmin(ORI) < 0.05 in at least one session with the standard object

configuration, indicating that these neurons were not exclusivelycoding for novelty. In contrast, none of the 29 MEC neuronsshowed elevated firing in response to novel objects, a result thatwas significantly different from LEC (Figure 5B; χ2 = 3.50, onetailed p = 0.0308).

Responses to misplaced objects were similar to novel objects.In LEC, 6/30 cells responded significantly to the misplaced object(z = 3.75, p = 8.9 × 10−5), whereas in MEC only 1 of 27 cells didso (z = –0.325, p = 0.63, n.s.). However, the difference betweenLEC and MEC was not significant for this comparison, per-haps due to the small numbers of cells (Figure 5C; χ2 = 2.15,one tailed p = 0.069). When returned to their original position,misplaced objects did not exert a stronger influence on LEC neu-rons than objects that had not been misplaced. Of the 10 LECneurons with pmin(ORI) < 0.05 in standard sessions immediatelyfollowing the misplaced object session, only two of the previously






FIGURE 5 | Responses of entorhinal cortical neurons to object

manipulations. (A) Object-responsive (units 1–2) and spatiallyselective (units 3–4) neurons in LEC. White circles mark the standardlocations of objects and stars represent the locations of novel(session 3) and misplaced (session 5) objects. Magenta lines connectthe standard (marked by x) and misplaced locations of objects in

session 5. The firing of units 3 and 4 did not depend on the animals’head direction being pointed toward the objects. (B) A significantlylarger proportion of LEC neurons respond to novel objects. (C) Asimilar trend is seen in response to misplaced objects. (D) Asignificantly larger proportion of LEC neurons shows object-relatedactivity in at least one session.

misplaced objects showed the lowest pORI among the four stan-dard objects. This proportion is not significantly different fromthe 25% expected by chance (test of proportions z = −0.3651,p = 0.35). Moreover, unit 2 in Figure 5 is one of these two units.This unit fires selectively at a standard location of one standardobject even before it is misplaced, as well as when the object is notat its standard position, thus showing that the object having beenmisplaced in the previous session is not the cause for this neuronto fire more at the misplaced object in the following session.

Three LEC neurons showed object–place conjunctiveresponses, in which they fired at an object only when it was movedto a new location and/or maintained the firing at the new locationwhen the object was returned to its standard location (Figure A2in Appendix, units 3, 23, 28; Weible et al., 2009). In total, a largerproportion of LEC neurons (26/61, or 43%) showed object-relatedactivity in at least one session (including standard as well as object-manipulation sessions) compared to MEC neurons (6/44, or 14%;χ2 = 8.81, one tailed p = 0.0017; Figure 5D). Object-responsiveneurons were recorded from multiple rats, and the proportion ofobject-responsive neurons within LEC and MEC was similar indifferent rats (Table 1).

LEC PLACE-LIKE ACTIVITY IN THE PRESENCE OF OBJECTSIn addition to the neurons that fired at objects, a number ofsuperficial LEC neurons had spatial firing fields away from objects(Figure 5A). Unit 3 had a small firing field at the south wall,whereas unit 4 had a strong field toward the middle of the box (but

away from any standard object location) and a weaker field nearone of the objects. In both cases, the fields were stable across ses-sions. These putative spatial fields were not present in the sessionswithout objects (Figure A1 in Appendix). Previous recordingsfrom LEC did not report these cells (Hargreaves et al., 2005; Yoga-narasimha et al., 2010), adding further evidence that objects maybe required for place-like activity to be present in LEC. We used aconservative set of criteria to identify a LEC cell as a putative place-related cell: (1) it had a statistically significant (p < 0.01) spatialinformation score >0.4 bits/spike; (2) the rate maps of the cellin consecutive sessions were highly correlated; and (3) its p(ORI)for all 4-objects-together was >0.4. Rate maps for the six neuronsclassified as putative place cells with fields away from objects areshown in Figure A2 in Appendix (Units 24, 43, 50, 66, 73, and 80).In light of this place-related activity contingent on the presenceof objects, two other neurons that were classified as object-related(unit 2 of Figure 5 and unit 74 of Figure A2 in Appendix) mayactually be place-related neurons that happened to have fields atthe objects.

HISTOLOGICAL LOCATIONS OF RECORDING SITESProjections from entorhinal cortex to hippocampus are topo-graphically organized, such that dorsal hippocampus receivesinputs from lateral LEC and dorso-caudal MEC while ventralhippocampus receives inputs from medial LEC and ventral MEC(Witter and Amaral, 2004). LEC neurons in the present study werespread over the entire lateral to medial extent of LEC, while MEC






Table 1 | Distribution of cells with putative object-related activity across rats.

Rat number

177 183 192 194 208 209 211

LEC Total units1 14 14 1 8 36 10 4

Units included in object-related analyses2 13 9 0 4 28 4 3

Putative object-responsive units 6 5 0 0 13 1 1

Putative place cells3 0 1 0 0 4 1 0

MEC Total Units1 5 40 11

Units included in object-related analyses2 3 33 8

Putative object-responsive units 0 5 1

1Units with good isolation and at least 50 spikes in at least 1 session.2Units with statistically significant (p < 0.01) spatial information score >0.25 bits/spike in at least 1 session.3This analysis was run only on LEC units.

neurons were confined to the dorso-caudal to intermediate parts ofMEC (Figure 6). Object-responsive neurons were detected alongthe entire medial–lateral extent of LEC. To confirm that the differ-ence in the sampling from LEC and MEC did not contribute to thehigher proportion of object-responsive neurons in LEC reportedin this study, we compared the proportions of object-responsiveneurons in the lateral half of LEC and the dorsal half of MEC(dorsolateral projection band), which project to the dorsal halfof hippocampus. The dorsolateral projection band of LEC hada higher proportion of object-responsive neurons than the corre-sponding region of MEC (LEC: 17/41, or 41%; MEC:4/27, or 15%;χ2 = 4.24, one tailed p = 0.02), confirming the main results of thispaper. It remains to be determined whether neurons in ventralMEC (which were not sampled here) show a higher proportion ofobject-responsiveness than neurons in dorso-caudal MEC.

Object-responsive neurons were found in both layer II and layerIII in LEC. Because of the small number of object-responsive neu-rons confirmed in each layer (6 in layer II, 13 in layer III, and7 near the layer II/III border) and the confound introduced bythe tendency for layer II neurons to be recorded on later daysthan layer III neurons on average, any quantitative differences inobject-responsiveness between the layers could not be estimated.

DISCUSSIONA longstanding debate on the nature of hippocampal encoding inrats has centered on whether hippocampal neurons are specializedfor encoding space, or whether they encode non-spatial variablesas well (O’Keefe, 1999; Shapiro and Eichenbaum, 1999). A consen-sus is emerging that these cells encode conjunctive representationsof individual items within a spatial location or context (O’Keefeand Nadel, 1978; Wiebe and Staubli, 1999; Moita et al., 2003;Komorowski et al., 2009; Manns and Eichenbaum, 2009). It isoften hypothesized that parallel processing streams convey spa-tial and non-spatial information to the hippocampus through theMEC and LEC, respectively (Burwell, 2000; Hargreaves et al., 2005;Knierim et al., 2006; Manns and Eichenbaum, 2006; Ranganath,2010; Yoganarasimha et al., 2010). Although single neurons inmonkey entorhinal cortex have been shown to respond to picturesof objects and their location on a monitor (Suzuki et al., 1997), andsingle neurons in rat LEC respond to objects (Zhu et al., 1995a)and odors (Young et al., 1997), these experiments were performed

under conditions that prevent a direct comparison to the spatialfiring properties of hippocampal place cells and MEC grid cells.The present results provide the first direct confirmation of thishypothesis in the context of the navigation/foraging tasks typicallyused to study spatial encoding in the hippocampus. In conjunc-tion with previous results that show strong spatial selectivity ofMEC neurons and little selectivity of LEC neurons in the absenceof objects (Hargreaves et al., 2005; Yoganarasimha et al., 2010),we show here a double dissociation between these areas, as LECneurons are much more strongly responsive to objects than areMEC neurons. In contrast, we observed a number of obvious gridcells in MEC (e.g., units 18, 23, and 45 in Figure A3 in Appendix),which were not present in LEC (Figures A1 and A2 in Appendix).

OBJECT SELECTIVITY IN LECThe nature of the object representations in LEC is not yet clear, asthere was a wide variety of responses to objects. A number of LECneurons fired at multiple objects, in multiple sessions (e.g., units 1,7, 23, 48, 58, and 71 in Figure A2 in Appendix), suggesting that theyare encoding object location or perhaps generalized attention toexternal landmarks. The latter suggestion might explain why manycells did not fire consistently at the same objects over sessions, asthe animal’s attention to a particular object may have varied acrosssessions. This interpretation is consistent with the idea that theLEC gates sensory input from perirhinal cortex to the hippocam-pus, allowing only behaviorally relevant or attended stimuli togain access to the hippocampus. Along these lines, Morris andFrey (1997) have suggested that the hippocampus automaticallyrecords only “attended” experience. In slices, electrical stimulationof perirhinal cortex does not activate LEC when the stimulationis restricted to perirhinal cortex, but it causes a strong activationof LEC when paired with stimulation of the amygdala (Kajiwaraet al., 2003; de Curtis and Pare, 2004). Thus, some type of saliencesignal may be critical to allow the sensory input from perirhi-nal cortex to drive LEC neurons. This notion is consistent withrecent results suggesting that the temporal or spatial stability ofhippocampal place fields may be modulated by changes in the ani-mal’s attention to external landmarks (Kentros et al., 2004; Muzzioet al., 2009; Fenton et al., 2010).

The foregoing discussion raises the possibility that LEC isencoding salience, rather than a salience-gated sensory input. That






FIGURE 6 | Single unit recording locations. The locations of units recordedfrom all rats are marked on coronal sections from one of the rats used in thisstudy. Red dots indicate the locations of units with pmin(ORI) < 0.05 shown inFigures A2 and A3 in Appendix. Green dots indicate the locations of LECputative place cells firing away from objects shown in Figure A2 in Appendix.

Blue dots show the locations of unclassified units. These sections are 470 μmapart (assuming a 15% histological shrinkage factor), and approximatelycorrespond to the following plates in the Paxinos and Watson (1998) rat brainatlas: section 1: plate 42 (bregma −5.6 mm) to section 10: plate 55 (bregma−8.8 mm). Scale bar is 1 mm.

is, the LEC neurons might fire whenever the animal pays attentionto what it perceives as a salient sensory stimulus without regardto the actual properties of the stimulus. This would imply thatsimultaneously recorded object-responsive neurons would showsimilar object selectivity (i.e., if neuron 1 prefers objects 1 and 4,so should neuron 2) and coordinated changes from session to ses-sion (i.e., if neuron 1 changes preference from object 1 in session1 to object 2 in session 2, so should neuron 2). We did not seesuch an effect in object-responsive neurons recorded simultane-ously (e.g., units 44 and 45; 47 and 48), making it unlikely that allLEC object-responsive neurons encode a salience signal withoutregards to the properties of the stimulus.

A few cells appeared to show some selectivity for subsets ofobjects (e.g., units 5, 12, 45, 47, 49, 64, and 74 in Figure A2 in

Appendix), suggesting that there may be a distributed, popula-tion code for object identity in the LEC. Such a code is consistentwith the weak object-identity signal identified in the hippocampusunder similar conditions of undirected exploration (Lenck-Santiniet al., 2005; Manns and Eichenbaum, 2009). Under other condi-tions in which the animals performed behavioral tasks related toindividual items, hippocampal cells showed greater selectivity forthe items (Wood et al., 1999; Komorowski et al., 2009). It is thuspossible that LEC neurons can show greater, more stable, individ-ual object selectivity under similar behavioral conditions. Further-more, even under the conditions of the present study (conditionsthat are similar to those employed in standard object-recognitiontasks, which measure differential investigation of novel objectsor misplaced objects in the absence of an overt behavioral task;






Mumby et al., 2002), there may be a very sparse representation ofindividual objects that we did not detect. Given the small num-ber of objects used in this study (each neuron was tested on fourstandard objects and one novel object), we cannot eliminate thepossibility that there are neurons in the LEC that respond to indi-vidual objects with high specificity, similar to the sparse code forindividual items reported in human medial temporal lobe neu-rons (Quiroga et al., 2005). Although these open questions do notdiminish the importance of the dissociation shown here betweenLEC and MEC in terms of overall object-responsiveness, they indi-cate future directions that are necessary to determine the precisenature of the object-related code in LEC and, subsequently, thenature of the computations performed by the hippocampus onthis input.

PLACE-RELATED ACTIVITY IN LEC IN THE PRESENCE OF OBJECTSThe consistent firing of LEC neurons at spatial locations in thearena away from the objects was an unexpected finding. Whileit is possible that this apparent place-related activity is merely arepresentation of “unidentified objects” such as odors and specificwalls, this is unlikely for the following reasons: 1. If the place-related cells were really just object-responsive cells responding to“unidentified objects,” they would be expected to show a sessionto session variability similar to the known object-responsive cells.One of the criteria for classifying a neuron as a putative place-related cell was that its intersession rate map correlation had to be>0.71. All 6 of the place-related cells had correlation coefficients>0.81, while the only object-responsive cell that had a correla-tion coefficient >0.81 was a cell (Cell 2 of Figure 5) that firedat a single object, and continued doing so at its standard loca-tion after the object was moved (and hence might be a putativeplace-related cell itself). 2. In the rare sessions in which the ratsproduced strong olfactory stimuli like urine or feces, the urine orfeces were removed and the area wiped with paper towels to spreadthe odor over a large area, making it less likely that a strong sen-sory stimulus with a very small spatial spread remained in the boxfrom session to session. 3. Previous experiments had conditionsthat should have given rise to similar responses to “unidentifiedobjects,” such as textures on the circular track, from which it wouldbe impossible to eliminate odors (Yoganarasimha et al., 2010), andwalls (Hargreaves et al., 2005; Yoganarasimha et al., 2010). Puta-tive place-related activity was not observed in LEC under theseexperimental conditions.

INFLUENCE OF OBJECTS ON FUNCTIONAL POPULATIONS IN MECNone of the MEC neurons with pmin(ORI) < 0.05 could be iden-tified as grid or border cells, even though some of the vertices ofhigh resolution grid cells were at or near the objects (e.g., units18, 23, and 45 in Figure A3 in Appendix). However, the influ-ence of objects on grid cells may not be limited to increasing theirpropensity to fire at the objects. For example, the objects, as land-marks, may play a role in anchoring the phase and orientation ofgrid cells. Individual object-manipulations, as done in the currentexperiment, are inadequate to test this possibility. The control ofgrid phase and orientation by objects can be tested in future studiesby coherent translation or rotation of all the objects relative to the

box (and distal landmarks) as has been done with hippocampalplace cells (Cressant et al., 1999).

IMPLICATIONS FOR THE COMPUTATIONAL FUNCTIONS OF LEC ANDMEC: EXTERNAL SENSORY INPUT VS. PATH INTEGRATIONA key finding of this study is the demonstration of both object-responsive neurons and putative place-responsive neurons in LECin the presence of local objects, suggesting that the distinctionbetween MEC and LEC may not be purely spatial vs. non-spatial.Rather, distinctions between these areas may be best describedin terms of computations based on internally based, path inte-gration mechanisms in MEC vs. computations based on the pro-cessing of external sensory input in LEC. The former requiresexternal sensory input to keep the path integration computa-tion stable relative to the external world, a function that may beperformed by boundary cells in MEC (Savelli et al., 2008; Sol-stad et al., 2008). This requirement is consistent with the smallamount of object-related activity seen in MEC in the presentstudy. In contrast, the LEC may primarily represent non-spatialinformation, such as objects, and may also create sparse, spatialrepresentations based on configurations of external, local land-marks (but not distal landmarks; Yoganarasimha et al., 2010).The three-dimensional objects used in the present study hada stronger influence on the activity of LEC neurons comparedto textures on the track (Yoganarasimha et al., 2010) or a cuecard in a box (Hargreaves et al., 2005), indicating that three-dimensionality and prominence of objects might be importantcorrelates of LEC activity. However, distal objects are not suffi-cient to give rise to the putative place cell like activity seen inthe presence of objects (Yoganarasimha et al., 2010), indicatingthat local three-dimensional objects may affect LEC processingdifferently from distal landmarks. Although the spatial selectiv-ity of LEC is low in the absence of local landmarks, the exter-nal sensory input from LEC may nonetheless be a contributingfactor that allows some hippocampal cells to retain place fieldswhen the MEC is lesioned or when grid cells are disrupted byabolishing the theta rhythm (Miller and Best, 1980; Van Cauteret al., 2008; Brandon et al., 2011; Koenig et al., 2011). Thus, thedouble dissociation between LEC and MEC is strong but notabsolute, which is expected given the anatomical connectivitybetween the LEC and MEC pathways and the feedback connec-tions from the hippocampus. Nonetheless, the clear differencesbetween these areas support the notion that they perform differ-ent computations and/or provide different types of informationto the hippocampus: external sensory input from LEC and path-integration-based, spatial input from MEC. The convergence ofthese two signals in the hippocampus might allow the formationof context-dependent, conjunctive representations of “what hap-pened where” that can be later retrieved during episodic memoryrecall.

AUTHOR CONTRIBUTIONSJames J. Knierim conceived the study. Sachin S. Deshmukh andJames J. Knierim designed the experiments and wrote the man-uscript. Sachin S. Deshmukh performed the experiments andanalyzed the data.






ACKNOWLEDGMENTSWe thank Geeta Rao for technical help; Dr. Francesco Savelli forhelpful discussions, comments on the manuscript, and assistance

with the mutual information analysis; and Drs. Kimberly Christ-ian and Raghav Rajan for comments on the manuscript. This workwas supported by NIH grant RO1 NS039456.

REFERENCESAggleton, J. P., and Brown, M. W. (1999).

Episodic memory, amnesia, and thehippocampal-anterior thalamic axis.Behav. Brain Sci. 22, 425–444.

Boccara, C. N., Sargolini, F., Thoresen,V.H., Solstad, T., Witter, M. P., Moser,E. I., and Moser, M. B. (2010). Gridcells in pre- and parasubiculum. Nat.Neurosci. 13, 987–994.

Brandon, M. P., Bogaard, A. R., Libby,C. P., Connerney, M. A., Gupta,K., and Hasselmo, M. E. (2011).Reduction of theta rhythm disso-ciates grid cell spatial periodicityfrom directional tuning. Science 332,595–599.

Burwell, R. D. (2000). The parahip-pocampal region: corticocorticalconnectivity. Ann. N. Y. Acad. Sci.911, 25–42.

Chen, L. L., Lin, L. H., Green, E. J.,Barnes, C. A., and McNaughton, B.L. (1994). Head-direction cells in therat posterior cortex. I. Anatomicaldistribution and behavioral modu-lation. Exp. Brain Res. 101, 8–23.

Cho, J., and Sharp, P. E. (2001). Headdirection, place, and movement cor-relates for cells in the rat retros-plenial cortex. Behav. Neurosci. 115,3–25.

Clayton,N. S., and Dickinson,A. (1998).Episodic-like memory during cacherecovery by scrub jays. Nature 395,272–274.

Cressant, A., Muller, R. U., and Poucet,B. (1999). Further study of the con-trol of place cell firing by intra-apparatus objects. Hippocampus 9,423–431.

de Curtis, M., and Pare, D. (2004).The rhinal cortices: a wall of inhibi-tion between the neocortex and thehippocampus. Prog. Neurobiol. 74,101–110.

Deshmukh, S. S., Yoganarasimha, D.,Voicu, H., and Knierim, J. J. (2010).Theta modulation in the medialand the lateral entorhinal cortices. J.Neurophysiol. 104, 994–1006.

Eichenbaum, H., and Fortin, N. J.(2005). Bridging the gap betweenbrain and behavior: cognitive andneural mechanisms of episodicmemory. J. Exp. Anal. Behav. 84,619–629.

Fenton, A. A., Lytton, W. W., Barry,J. M., Lenck-Santini, P. P., Zinyuk,L. E., Kubik, S., Bures, J., Poucet,B., Muller, R. U., and Olypher, A.V. (2010). Attention-like modulationof hippocampus place cell discharge.J. Neurosci. 30, 4613–4625.

Frank, L. M., Brown, E. N., and Wil-son, M. A. (2001). A comparison ofthe firing properties of putative exci-tatory and inhibitory neurons fromCA1 and the entorhinal cortex. J.Neurophysiol. 86, 2029–2040.

Hafting, T., Fyhn, M., Molden, S., Moser,M. B., and Moser, E. I. (2005).Microstructure of a spatial map inthe entorhinal cortex. Nature 436,801–806.

Hampson, R. E., Simeral, J. D., andDeadwyler, S. A. (1999). Distri-bution of spatial and nonspatialinformation in dorsal hippocampus.Nature 402, 610–614.

Hargreaves, E. L., Rao, G., Lee, I., andKnierim, J. J. (2005). Major dis-sociation between medial and lat-eral entorhinal input to dorsal hip-pocampus. Science 308, 1792–1794.

Hargreaves, E. L., Yoganarasimha, D.,and Knierim, J. J. (2007). Cohesive-ness of spatial and directional rep-resentations recorded from neuralensembles in the anterior thalamus,parasubiculum, medial entorhinalcortex, and hippocampus. Hip-pocampus 17, 826–841.

Kajiwara, R., Takashima, I., Mimura, Y.,Witter, M. P., and Iijima, T. (2003).Amygdala input promotes spreadof excitatory neural activity fromperirhinal cortex to the entorhinal-hippocampal circuit. J. Neurophys-iol. 89, 2176–2184.

Kentros, C. G., Agnihotri, N. T., Streater,S., Hawkins, R. D., and Kandel, E. R.(2004). Increased attention to spa-tial context increases both place fieldstability and spatial memory. Neuron42, 283–295.

Knierim, J. J. (2006). Neural rep-resentations of location outsidethe hippocampus. Learn. Mem. 13,405–415.

Knierim, J. J., and Hamilton, D. A.(2011). Framing spatial cognition:neural representations of proxi-mal and distal frames of refer-ence and their roles in navigation.Physiol. Rev. 91. doi:10.1152/phys-rev.00021.2010. (in press).

Knierim, J. J., Lee, I., and Hargreaves,E. L. (2006). Hippocampal placecells: parallel input streams, subre-gional processing, and implicationsfor episodic memory. Hippocampus16, 755–764.

Koenig, J., Linder, A. N., Leutgeb, J. K.,and Leutgeb, S. (2011). The spatialperiodicity of grid cells is not sus-tained during reduced theta oscilla-tions. Science 332, 592–595.

Komorowski, R. W., Manns, J. R., andEichenbaum,H. (2009). Robust con-junctive item-place coding by hip-pocampal neurons parallels learningwhat happens where. J. Neurosci. 29,9918–9929.

Lenck-Santini, P. P., Rivard, B., Muller,R. U., and Poucet, B. (2005).Study of CA1 place cell activityand exploratory behavior followingspatial and nonspatial changes inthe environment. Hippocampus 15,356–369.

Lever, C., Burton, S., Jeewajee, A.,O’Keefe, J., and Burgess, N. (2009).Boundary vector cells in the subicu-lum of the hippocampal formation.J. Neurosci. 29, 9771–9777.

Manns, J. R., and Eichenbaum, H.(2006). Evolution of declarativememory. Hippocampus 16, 795–808.

Manns, J. R., and Eichenbaum, H.(2009). A cognitive map for objectmemory in the hippocampus. Learn.Mem. 16, 616–624.

Miller, V. M., and Best, P. J. (1980). Spa-tial correlates of hippocampal unitactivity are altered by lesions of thefornix and entorhinal cortex. BrainRes. 194, 311–323.

Moita, M. A., Rosis, S., Zhou, Y.,LeDoux, J. E., and Blair, H. T. (2003).Hippocampal place cells acquirelocation-specific responses to theconditioned stimulus during audi-tory fear conditioning. Neuron 37,485–497.

Morris, R. G., and Frey, U. (1997). Hip-pocampal synaptic plasticity: role inspatial learning or the automaticrecording of attended experience?Philos. Trans. R. Soc. Lond. B Biol.Sci. 352, 1489–1503.

Mumby, D. G., Gaskin, S., Glenn, M.J., Schramek, T. E., and Lehmann,H. (2002). Hippocampal damageand exploratory preferences in rats:memory for objects, places, and con-texts. Learn. Mem. 9, 49–57.

Murray, E. A., Bussey, T. J., and Sak-sida, L. M. (2007). Visual perceptionand memory: a new view of medialtemporal lobe function in primatesand rodents. Annu. Rev. Neurosci. 30,99–122.

Muzzio, I. A., Levita, L., Kulkarni,J., Monaco, J., Kentros, C., Stead,M., Abbott, L. F., and Kandel, E.R. (2009). Attention enhances theretrieval and stability of visuospa-tial and olfactory representationsin the dorsal hippocampus. PLoSBiol. 7, e1000140. doi:10.1371/jour-nal.pbio.1000140

O’Keefe, J. (1976). Place units in the hip-pocampus of the freely moving rat.Exp. Neurol. 51, 78–109.

O’Keefe, J. (1999). Do hippocampalpyramidal cells signal non-spatialas well as spatial information? Hip-pocampus 9, 352–364.

O’Keefe, J., and Nadel, L. (1978). TheHippocampus as a Cognitive Map.Oxford: Clarendon Press.

Paxinos, G., and Watson, C. (1998). TheRat Brain in Stereotaxic Coordinates.San Diego: Academic Press.

Quiroga, R. Q., Reddy, L., Kreiman, G.,Koch, C., and Fried, I. (2005). Invari-ant visual representation by singleneurons in the human brain. Nature435, 1102–1107.

Ranganath, C. (2010). A unified frame-work for the functional organi-zation of the medial temporallobes and the phenomenology ofepisodic memory. Hippocampus 20,1263–1290.

Rivard, B., Li, Y., Lenck-Santini, P. P.,Poucet, B., and Muller, R. U. (2004).Representation of objects in spaceby two classes of hippocampal pyra-midal cells. J. Gen. Physiol. 124,9–25.

Sargolini, F., Fyhn, M., Hafting, T.,McNaughton, B. L., Witter, M. P.,Moser, M. B., and Moser, E. I.(2006). Conjunctive representationof position, direction, and velocityin entorhinal cortex. Science 312,758–762.

Savelli, F., Yoganarasimha, D., andKnierim, J. J. (2008). Influenceof boundary removal on the spa-tial representations of the medialentorhinal cortex. Hippocampus 18,1270–1282.

Scoville, W. B., and Milner, B. (1957).Loss of recent memory after bilat-eral hippocampal lesions. J. Neurol.Neurosurg. Psychiatr. 20, 11–21.

Shapiro, M. L., and Eichenbaum, H.(1999). Hippocampus as a memorymap: synaptic plasticity and mem-ory encoding by hippocampal neu-rons. Hippocampus 9, 365–384.

Sharp, P. E. (1997). Subicular cells gen-erate similar spatial firing patterns intwo geometrically and visually dis-tinctive environments: comparisonwith hippocampal place cells. Behav.Brain Res. 85, 71–92.

Skaggs, W. E., McNaughton, B. L.,Wilson, M. A., and Barnes, C. A.(1996). Theta phase precession inhippocampal neuronal populationsand the compression of temporalsequences. Hippocampus 6, 149–172.


http://dx.doi.org/10.1152/physrev.00021.2010. (in press)

http://dx.doi.org/10.1371/journal.pbio.1000140





Solstad, T., Boccara, C. N., Kropff, E.,Moser, M. B., and Moser, E. I. (2008).Representation of geometric bordersin the entorhinal cortex. Science 322,1865–1868.

Squire, L. R., Stark, C. E., and Clark, R.E. (2004). The medial temporal lobe.Annu. Rev. Neurosci. 27, 279–306.

Suzuki, W. A., Miller, E. K., and Desi-mone, R. (1997). Object and placememory in the macaque entorhi-nal cortex. J. Neurophysiol. 78,1062–1081.

Taube, J. S. (1995). Place cells recordedin the parasubiculum of freelymoving rats. Hippocampus 5,569–583.

Van Cauter, T., Poucet, B., and Save, E.(2008). Unstable CA1 place cell rep-resentation in rats with entorhinalcortex lesions. Eur. J. Neurosci. 27,1933–1946.

Vargha-Khadem, F., Gadian, D. G.,Watkins, K. E., Connelly, A., VanPaesschen, W., and Mishkin, M.(1997). Differential effects of earlyhippocampal pathology on episodic

and semantic memory. Science 277,376–380.

Wan, H., Aggleton, J. P., and Brown, M.W. (1999). Different contributionsof the hippocampus and perirhi-nal cortex to recognition memory. J.Neurosci. 19, 1142–1148.

Weible, A. P., Rowland, D. C., Pang,R., and Kentros, C. (2009). Neuralcorrelates of novel object and novellocation recognition behavior in themouse anterior cingulate cortex. J.Neurophysiol. 102, 2055–2068.

Wiebe, S. P., and Staubli, U. V. (1999).Dynamic filtering of recognitionmemory codes in the hippocampus.J. Neurosci. 19, 10562–10574.

Wiener, S. I., Paul, C. A., and Eichen-baum, H. (1989). Spatial and behav-ioral correlates of hippocampalneuronal activity. J. Neurosci. 9,2737–2763.

Witter, M. P., and Amaral, D. G. (2004).“Hippocampal formation,” in TheRat Nervous System, 3rd Edn, ed.G. Paxinos (Amsterdam: Elsevier),635–704.

Wood, E. R., Dudchenko, P. A., andEichenbaum, H. (1999). The globalrecord of memory in hippocam-pal neuronal activity. Nature 397,613–616.

Yoganarasimha, D., Rao, G., andKnierim, J. J. (2010). Lateral entorhi-nal neurons are not spatially selec-tive in cue-rich environments. Hip-pocampus. doi:10.1002/hipo.20839.[Epub ahead of print].

Young, B. J., Otto, T., Fox, G. D., andEichenbaum, H. (1997). Memoryrepresentation within the parahip-pocampal region. J. Neurosci. 17,5183–5195.

Zhu, X. O., Brown, M. W., and Aggleton,J. P. (1995a). Neuronal signallingof information important to visualrecognition memory in rat rhinaland neighbouring cortices. Eur. J.Neurosci. 7, 753–765.

Zhu, X. O., Brown, M. W., McCabe,B. J., and Aggleton, J. P. (1995b).Effects of the novelty or familiar-ity of visual stimuli on the expres-sion of the immediate early gene

c-fos in rat brain. Neuroscience 69,821–829.

Conflict of Interest Statement: Theauthors declare that the research wasconducted in the absence of anycommercial or financial relationshipsthat could be construed as a potentialconflict of interest.

Received: 22 August 2011; accepted: 03October 2011; published online: 28 Octo-ber 2011.Citation: Deshmukh SS and Knierim JJ(2011) Representation of non-spatial andspatial information in the lateral entorhi-nal cortex. Front. Behav. Neurosci. 5:69.doi: 10.3389/fnbeh.2011.00069Copyright © 2011 Deshmukh andKnierim. This is an open-access arti-cle subject to a non-exclusive licensebetween the authors and Frontiers MediaSA, which permits use, distribution andreproduction in other forums, providedthe original authors and source are cred-ited and other Frontiers conditions arecomplied with.


http://dx.doi.org/10.1002/hipo.20839

http://dx.doi.org/10.3389/fnbeh.2011.00069





APPENDIX

FIGURE A1 | Continued












FIGURE A1 | Spatial information in LEC is higher in an open field

containing objects compared to an empty field. To test whether the higherspatial information recorded with objects in the present study, compared toprevious studies (Hargreaves et al., 2005; Yoganarasimha et al., 2010), was anartifact of uncontrolled differences between the studies, 25 LEC units in tworats were recorded in sessions with (session 1) and without (session 7)objects. Because the LEC firing patterns might have been affected by a priorhistory of the presence of objects, the session without objects wasconducted in a similar box in a different room to minimize such a confound.(A) Firing rate maps of the neurons with higher spatial information scores inthe presence of objects than in the absence of objects, sorted in decreasingorder of the difference. Note that a number of cells (units 1, 3, 4, 5, 7, and 8)had highly localized, high-rate firing fields in the presence of objects butweaker, more diffuse firing in the absence of objects. Two cells (units 2 and 6)fired at higher rates in the session without objects, but they fired alongmultiple walls, not in restricted locations. Similar activity along walls has beenshown previously (Hargreaves et al., 2005). Peak firing rate (pk, Hz), spatialinformation score (i, bits/spike) and probability of getting the information scoreby chance (p) are shown at the top of each plot. Unlike the firing rate mapsshown elsewhere in the paper, which were autoscaled between 0 Hz andmaximum firing rates within the individual rate maps, the firing rate maps in(A,B) were scaled such that blue corresponds to 0 Hz while red correspondsto the larger of the peak firing rates in the with- and without-object sessionsfor the given neuron. This cross-session scaling makes it easy to see rateremapping as well as the locations of firing fields. Note that in some cases,the scaling masks low-rate firing that still results in moderate spatialinformation scores (e.g., units 1 and 4 without objects show informationscores of 0.56 and 0.49, respectively, although the peak firing rates andinformation scores are less than they are with objects). (B) Firing rate maps ofthe neurons with lower spatial information scores in the presence of objectsthan in the absence of objects, sorted in decreasing order of the difference.Note the lack of a pronounced difference in spatial firing selectivity betweenthe with-object and without-object sessions in these cells, in contrast withthe numerous examples in (A). This contrast argues strongly against ageneralized “remapping” interpretation of these data, as such an explanationwould predict the number of cells having higher spatial information in the

with-objects session to be approximately equal to the number of cells havinghigher information score in the without-objects environments. On average,the firing rate maps without objects in (A,B) resemble those shown in priorreports of LEC non-spatial firing (Hargreaves et al., 2005; Yoganarasimhaet al., 2010), with none of the cells showing robust, highly localized firing,indicating that the increased responsiveness when objects are present is notdue to a generalized increase in spatial selectivity in the present study. Bothrats included in this analysis were trained extensively in the environment withobjects, and the last 2–3 days of training included one foraging session in theenvironment without objects. The experiments were run over multiple days,making the second room more familiar over time. Furthermore, because theprior studies recorded from highly familiar environments and showed poorspatial selectivity in LEC, the similar lack of spatial selectivity without objectsin the present study is unlikely a result of the relative novelty of theenvironment without objects. (C) Comparison of spatial information scores ofLEC neurons in the presence and absence of objects. Red lines connectspatial information scores in the presence (+) and in the absence (−) ofobjects for neurons that showed higher spatial information scores in thepresence of objects than in the absence of objects, shown in (A). Blue linesconnect spatial information scores for neurons that showed lower spatialinformation in the presence of objects than in the absence of objects, shownin (B). Visually, the slopes of the red lines are on average greater than theslopes of the blue lines, indicating that a number of cells that had high spatialinformation in the presence of objects lost this tuning in the absence ofobjects. There were no neurons that had high spatial information in theabsence of objects and much lower information in the presence of objects(i.e., there are no blue lines with a steep slope), arguing against a generalremapping explanation for differences between the environments. Across allneurons, the spatial information scores were significantly higher in thepresence of objects than in the absence of objects (with-objectsmedian = 0.33 bits/spike, without-objects median = 0.24 bits/spike; Wilcoxonsigned rank test, p = 0.04).This difference was even more significant whenonly the 23 cells with significant information scores (p < 0.01) in at least oneof the two sessions were included (with-objects median = 0.33 bits/spike,without-objects median = 0.22 bits/spike; Wilcoxon signed rank test,p = 0.017).




























































FIGURE A2 | Rate maps of all well-isolated neurons recorded from LEC

with >50 spikes in the given session. All rate maps shown were smoothedusing adaptive binning (Skaggs et al., 1996), but unsmoothed rate maps wereused for the analyses of object-responsiveness. The positions of familiarobjects in their standard locations are marked with white circles, whilemisplaced and novel objects (in sessions 3/5) are marked with stars.Magenta lines connect the standard and new locations of misplaced objects.The peak firing rate (pk, spikes/second), spatial information score (i,bits/spike), and probability of obtaining the information score by chance (p)are shown at the top of each rate map. For standard object sessions(sessions 1, 2, 4, and 6), pmin (ORI) is shown at the bottom of each rate map[marked as p(ORI)]. For novel or misplaced object sessions (sessions 3 and5), the probability of getting higher firing at the novel [p(n)] or misplaced[p(m)] object is shown. Font colors for these probabilities are red when theyare <0.05 and when the rate maps have a spatial information score>0.25 bits/spike that is statistically significant at the p < 0.01 level. These arethe neurons included in the χ2 statistics reported in the paper. While manyLEC neurons showing object-related firing showed a lot of session to sessionvariability in terms of firing rates as well as the subsets of objects they firedat (e.g., units 1, 7, 23, 48, 58, 71, and 86), some neurons repeatedly fired atthe same subset of objects over multiple sessions (e.g., units 5, 12, 45, 47,49, 64, and 74). Thus, a subset of LEC neurons may convey information aboutobject identity in a distributed, population code. A number of LEC cells wereidentified as putative place cells, using the conservative criteria of a highspatial information score >0.4 bits/spike, high session to session stability,and a low probability of object-related activity (see Materials and Methods).These neurons are marked to the left of session 1 with a “Place” label (units24, 43, 50, 66, 73, 80). There are other putative place cells visible, whichmight have failed on one on more criteria, but which show distinct firing fieldsaway from the objects (e.g., units 2, 3, 9, 11, 53). Units 5 and 74, which fire atsingle objects, may also be place-related. This interpretation is supported bythe continued firing of these neurons in the same locations when the objectswere moved to different locations. Three LEC neurons (units 3, 23, 28) showobject–place conjunctive responses. Unit 3 fires at a misplaced object insession 3. It does not fire at this object in the other three sessions when the

object is in its standard position. Unit 23 fires at the misplace location of anobject (and weakly at the position where it used to be) in session 3. Itcontinues firing at this new location when the object is moved back to itsstandard location. It does not fire at the object in sessions 2 and 4, when theobject is in its standard position. Unit 28 fires at the misplaced locations ofobjects in session 3 [although p(m) is greater than 0.05] and continues to fireat the new locations in session 4, after the objects have been moved back totheir standard positions. All these responses cannot be explained as purelyobject-related or purely space-related activity, but are correlated to object andspace. Similar activity has previously been shown in hippocampus (O’Keefe,1976) and cingulate cortex (Weible et al., 2009). In addition to the probabilitiesshown with ratemaps for all other neurons, ratemaps for units 23 and 28 alsoshow p(m3), which is the probability in session 4 that the ORI at themisplaced location (where the object used to be in session 3) can beobtained by chance. A variety of considerations led to the exclusion of someof the neurons recorded in some of the sessions from the analyses. If a cellfired less than 50 spikes in a given session, or showed a drop in waveformsignal-to-noise in a given session so as to make its isolation qualityunacceptable, the cell was eliminated from the analysis, and its rate map notshown here. In addition, sessions in which the rat foraged poorly wereexcluded. All the decisions about excluding cells/sessions were madewithout regard to spatial firing characteristics of the neurons in the givensession. Behavioral biases of the rats (e.g., their tendency to run counterclockwise along the periphery, and approach objects in a stereotyped manneron the way in from the periphery) lead to unoccupied pixels near objectsseen in many of the rate maps, especially in the sessions that the rats did notforage very well. These pixels often tend to be on the southwest side of theobjects. Object-related calculations were performed only if there were atleast 15 occupied pixels within a 5-pixel radius of the given object in aparticular session. The reasons for the behavioral bias are not known, butbecause this bias was common to the rats with LEC as well as MECrecordings, it should not affect the comparison of spatial information as wellas object-related activity between the two areas. The behavioral correlates ofthe rats’ interactions with the objects, beyond the purview of the currentstudy, will shed light on the nature of object-related activity seen here.










































FIGURE A3 | Rate maps of all well-isolated neurons recorded from MEC with >50 spikes in the given session. See Figure A2 caption for details.





Documents

Representation of non-spatial and spatial information in ...Representation of non-spatial and spatial information in ... like” memory in animals – memory that requires an integration